Predicting Tensile Strength of Limestone Rocks from Experimental Data of Physical Properties: Statistical Correlations and Machine-Learning Models

Limestone rocks are increasingly being excavated for infrastructures developments and mining projects, and for extracting construction materials. This calls for data and models to be available for practicing engineers to predict mechanical properties of limestone from basic rock indices. This paper aims at addressing this gap by developing statistical models to predict the tensile strength of local limestone, from basic rock indices. The methodology includes experimental work to obtain data of local limestone properties, and statistical analysis to correlate the tensile strength with various rock properties. The experimental study used 100 cylindrical cores for index-property tests and compressive-strength tests, 100 disks for Brazilian tensile-strength strength tests, and 100 prisms for flexural strength tests. Additionally, rock indices were obtained for these samples, including dry density, P-wave velocity, hardness, porosity, specific gravity, water absorption, and point-load strength index, in addition to compressive strength. Statistical analysis includes simple-regression models for interpretable prediction, and machine-learning models used to evaluate nonlinear effects and predictive performance. The machine-learning models achieved repeated tenfold cross-validated R² of about 0.88 for the Brazilian tensile strength and 0.62 for the flexural tensile strength, and provided distribution-free conformal-prediction intervals for uncertainty-aware use. Results of experimental work are presented for the index properties, the compressive strength, and the tensile strength (Brazilian and flexural). Outcomes of the regression analysis indicated that TB and TF are strongly correlated with the stated index properties, and also TB correlated well with TF. The results compare well with models available in the literature. It can be concluded that the presented models enable the estimation of tensile strength of limestone from index properties that can be obtained from simple, quick, inexpensive, and non-destructive tests.